import sys
import os
from os import path
from sklearn.preprocessing import MinMaxScaler
sys.path.append(path.join(path.dirname(__file__), '../'))
sys.path.insert(0, "../")
import sys
sys.path.append(os.getcwd())
import numpy as np
from hwer import MultiCategoricalEmbedding, FeatureSet, Feature, FeatureType
f1 = Feature(
"f1", FeatureType.MULTI_CATEGORICAL,
[["a", "b"], ["b"], ["c", "b"], ["a"], ["a", "c"], ["a", "b"], ["b"]])
f2 = Feature("f2", FeatureType.NUMERIC, [1.0, 3.0, 2.0, 4.0, 5.0, 6.0, 2.0])
fs = FeatureSet([f2])
cs = MultiCategoricalEmbedding(4, True)
p = cs.fit_transform(f1, target=fs)
print(p)
# without target
cs = MultiCategoricalEmbedding(4, True)
p = cs.fit_transform(f1)
print(p)
import sys
import os
from os import path
from sklearn.preprocessing import MinMaxScaler
sys.path.append(path.join(path.dirname(__file__), '../'))
sys.path.insert(0, "../")
import sys
sys.path.append(os.getcwd())
import numpy as np
from hwer import CategoricalEmbedding, FeatureSet, Feature, FeatureType
f1 = Feature("f1", FeatureType.CATEGORICAL, ["a","b","c","b","c","c","a"])
f2 = Feature("f2", FeatureType.NUMERIC, [1.0,3.0,2.0,4.0,5.0,6.0,2.0])
fs = FeatureSet([f2])
cs = CategoricalEmbedding(4, True)
p = cs.fit_transform(f1, target=fs)
print(p)
cs = CategoricalEmbedding(4, True)
p = cs.fit_transform(f1)
print(p)
def test_once(train_affinities, validation_affinities, items, capabilities=["svdpp", "resnet", "content", "triplet", "implicit"]):
embedding_mapper = {}
embedding_mapper['gender'] = CategoricalEmbedding(n_dims=2)
embedding_mapper['age'] = CategoricalEmbedding(n_dims=2)
embedding_mapper['occupation'] = CategoricalEmbedding(n_dims=4*kfold_multiplier)
embedding_mapper['zip'] = CategoricalEmbedding(n_dims=2*kfold_multiplier)
embedding_mapper['text'] = FlairGlove100AndBytePairEmbedding()
embedding_mapper['numeric'] = NumericEmbedding(4*kfold_multiplier)
embedding_mapper['genres'] = MultiCategoricalEmbedding(n_dims=4*kfold_multiplier)
u1 = Feature(feature_name="gender", feature_type=FeatureType.CATEGORICAL, values=users.gender.values)
u2 = Feature(feature_name="age", feature_type=FeatureType.CATEGORICAL, values=users.age.astype(str).values)
u3 = Feature(feature_name="occupation", feature_type=FeatureType.CATEGORICAL,
values=users.occupation.astype(str).values)
u4 = Feature(feature_name="zip", feature_type=FeatureType.CATEGORICAL, values=users.zip.astype(str).values)
user_data = FeatureSet([u1, u2, u3, u4])
i1 = Feature(feature_name="text", feature_type=FeatureType.STR, values=movies.text.values)
i2 = Feature(feature_name="genres", feature_type=FeatureType.MULTI_CATEGORICAL, values=movies.genres.values)
i3 = Feature(feature_name="numeric", feature_type=FeatureType.NUMERIC,
values=movies[["title_length", "overview_length", "runtime"]].values)
item_data = FeatureSet([i1, i2, i3])
kwargs = {}
kwargs['user_data'] = user_data
kwargs['item_data'] = item_data
kwargs["hyperparameters"] = copy.deepcopy(hyperparameters)
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"]["use_svd"] = False
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"]["use_resnet"] = False
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"]["resnet_content_each_layer"] = False
kwargs["hyperparameters"]['collaborative_params'][
"use_triplet"] = False
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"][
"use_implicit"] = False
if "svdpp" in capabilities:
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"]["use_svd"] = True
if "resnet" in capabilities:
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"]["use_resnet"] = True
if "content" in capabilities:
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"][
"resnet_content_each_layer"] = True
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"]["use_resnet"] = True
if "triplet" in capabilities:
kwargs["hyperparameters"]['collaborative_params'][
"use_triplet"] = True
if "implicit" in capabilities:
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"][
"use_implicit"] = True
if "dnn" in capabilities or "resnet" in capabilities:
kwargs["hyperparameters"]['collaborative_params']["prediction_network_params"][
"use_dnn"] = True
recsys = HybridRecommenderSVDpp(embedding_mapper=embedding_mapper, knn_params=None, rating_scale=(1, 5),
n_content_dims=32 * kfold_multiplier,
n_collaborative_dims=32 * kfold_multiplier)
start = time.time()
user_vectors, item_vectors = recsys.fit(users.user_id.values, movies.movie_id.values,
train_affinities, **kwargs)
# cos_sims = []
# for i in range(len(item_vectors)):
# cos_sims.append([])
# for j in range(len(item_vectors)):
# sim = cos_sim(item_vectors[i], item_vectors[j])
# cos_sims[i].append(sim)
# cos_sims = np.array(cos_sims)
# print(cos_sims.min(), cos_sims.max(), cos_sims.mean())
end = time.time()
total_time = end - start
predictions, actuals, rmse, mae = get_prediction_details(recsys, validation_affinities)
_, _, train_rmse, train_mae = get_prediction_details(recsys, train_affinities)
print("hybrid-" + "_".join(capabilities), ": ", rmse, mae, train_rmse, train_mae)
ex_ee = extraction_efficiency(recsys, train_affinities, validation_affinities, model_get_topk, items)
if enable_error_analysis:
error_df = pd.DataFrame({"errors": actuals - predictions, "actuals": actuals, "predictions": predictions})
error_analysis(error_df, "Hybrid")
results = [{"algo":"hybrid-" + "_".join(capabilities), "rmse": rmse, "mae": mae,
"map": ex_ee["map"], "retrieval_time": ex_ee["retrieval_time"],
"train_rmse": train_rmse, "train_mae": train_mae, "time": total_time}]
return recsys, results, predictions, actuals
embedding_mapper['gender'] = CategoricalEmbedding(n_dims=2)
embedding_mapper['age'] = CategoricalEmbedding(n_dims=2)
embedding_mapper['occupation'] = CategoricalEmbedding(n_dims=2)
embedding_mapper['zip'] = CategoricalEmbedding(n_dims=8)
embedding_mapper['title'] = FlairGlove100AndBytePairEmbedding()
embedding_mapper['genres'] = MultiCategoricalEmbedding(n_dims=16)
recsys = ContentRecommendation(embedding_mapper=embedding_mapper,
knn_params=None,
n_output_dims=64)
kwargs = {'user_item_affinities': user_item_affinities}
u1 = Feature(feature_name="gender",
feature_type=FeatureType.CATEGORICAL,
values=users.gender.values)
u2 = Feature(feature_name="age",
feature_type=FeatureType.CATEGORICAL,
values=users.age.astype(str).values)
u3 = Feature(feature_name="occupation",
feature_type=FeatureType.CATEGORICAL,
values=users.occupation.astype(str).values)
u4 = Feature(feature_name="zip",
feature_type=FeatureType.CATEGORICAL,
values=users.zip.astype(str).values)
user_data = FeatureSet([u1, u2, u3, u4])
i1 = Feature(feature_name="title",
feature_type=FeatureType.STR,
values=movies.title.values)
from os import path
from sklearn.preprocessing import MinMaxScaler
sys.path.append(path.join(path.dirname(__file__), '../'))
sys.path.insert(0, "../")
import sys
sys.path.append(os.getcwd())
import numpy as np
from gensim.test.utils import common_texts
from hwer import FasttextEmbedding, Feature, FeatureType
ft = FasttextEmbedding(
32,
fasttext_file=
"/Users/ahemf/mygit/Hybrid-Weighted-Embedding-Recommender/hwer/fasttext.bin"
)
text = list(map(lambda x: " ".join(x), common_texts))
f1 = Feature("text", FeatureType.STR, text)
print(ft.fit_transform(f1))
print("=" * 40)
ft = FasttextEmbedding(4, )
f1 = Feature("text", FeatureType.STR, text)
print(ft.fit_transform(f1))
print(ft.fit_transform(f1).shape)
user_item_affinities = []
user_embeddings = []
for i, user in enumerate(user_ids):
user_embeddings.append(np.average(i1_15[i*items_per_user:(i+1)*items_per_user], axis=0,))
for j in range(i*items_per_user, (i+1)*items_per_user):
user_item_affinities.append((user, item_ids[j], 3))
user_embeddings = np.vstack(user_embeddings)
actual_embeddings = np.concatenate((user_embeddings, i1_15))
embedding_mapper = {}
embedding_mapper['numeric'] = NumericEmbedding(n_dims=3)
f = Feature("numeric", FeatureType.NUMERIC, i1_15)
item_data = FeatureSet([f])
kwargs = {'item_data': item_data}
recsys = ContentRecommendation(embedding_mapper=embedding_mapper, knn_params=None, n_output_dims=2, rating_scale=(1, 5))
_ = recsys.fit(user_ids, item_ids,
user_item_affinities, **kwargs)
all_entities = list(zip(user_ids, [EntityType.USER]*len(user_ids))) +\
list(zip(item_ids, [EntityType.ITEM]*len(item_ids)))
embeddings = recsys.get_embeddings(all_entities)
all_entities = list(zip(user_ids, [EntityType.USER]*len(user_ids), [5]*len(user_ids))) +\
list(zip(item_ids, [EntityType.ITEM]*len(item_ids), [1]*len(item_ids)))
sys.path.append(path.join(path.dirname(__file__), '../'))
sys.path.insert(0, "../")
import sys
sys.path.append(os.getcwd())
import numpy as np
from gensim.test.utils import common_texts
from hwer import FlairGlove100AndBytePairEmbedding,FlairGlove100Embedding, Feature, FeatureType
# text = list(map(lambda x: " ".join(x), common_texts))
#
# f1 = Feature("text", FeatureType.STR, text)
#
# flair1 = FlairGlove100Embedding()
# print(flair1.fit_transform(f1))
#
# flair2 = FlairGlove100AndBytePairEmbedding()
# print(flair2.fit_transform(f1))
#
# print(flair1.fit_transform(f1).shape)
# print(flair2.fit_transform(f1).shape)
#
f1 = Feature("text", FeatureType.STR, ["eifjcchchbnikfncbcntnhbvthnrbjiechcrbinucknb"])
flair1 = FlairGlove100Embedding()
print(flair1.fit_transform(f1))
import sys
import os
from os import path
from sklearn.preprocessing import MinMaxScaler
sys.path.append(path.join(path.dirname(__file__), '../'))
sys.path.insert(0, "../")
import sys
sys.path.append(os.getcwd())
import numpy as np
from hwer import CategoricalEmbedding, FeatureSet, Feature, NumericEmbedding, FeatureType
f1 = Feature("f1", FeatureType.NUMERIC,
[1.2, 0.1, 2.2, 4.1, 5.0, 6.1, 2.1, 5.0])
f2 = Feature("f2", FeatureType.NUMERIC,
[0.7, 3.0, 2.0, 4.0, 5.0, 6.0, 7.0, 5.0])
new_vals = list(zip(f1.values, f2.values))
print(new_vals)
f = Feature("f1", FeatureType.NUMERIC, new_vals)
ns = NumericEmbedding(4)
print(ns.fit_transform(f))
ns = NumericEmbedding(4)
print(ns.fit_transform(f1))